Humanity's Hard Problems
Image: PixaBay 
MODULE 1.0
Humanity's Hard Problems
 

Humanity is facing many existential problems like Global Warming and Climate Change, the need for alternative energy sources and the emergence of pandemics like COVID-19. These problems are extremely complex and need deep scientific insights and the capability to inference across large amounts of even loosely related knowledge. This need might exceed the human cognitive capacity at large.

In many discussions on the dangers of Super Intelligence, the doom scenarios have been described by people like Elon Musk, Bill Gates and the late Stephen Hawking. In stark contrast, it appears that we actually NEED Super Intelligent machines to help us solve these existential threats to humanity.

 

The dangers that are ascribed to some unknown and undefined future Super Intelligence, are completely absent in the ASTRID system. The ASTRID system is safe, even when it becomes Super Intelligent because we know in what way it can become Super Intelligent. The more intelligent the ASTRID system becomes, the more it understands why some things are good and, more importantly, why certain things are bad.

Here at MIND|CONSTRUCT we already started the Global Societal Risks Initiative (GSRI), aimed at building and training a special ASTRID version for this specific goal. Currently, this is an internal project that is in the pilot stage. Over the course of the coming year we hope to extend this project outside our company and to work with other entities in the field to make this reality.

© 2021 MIND|CONSTRUCT  
Deep Inference
Image: Athena 
MODULE 1.1
Deep Inference
 

Deep Inference is one of the main concepts when we talk about the strong points of the ASTRID system and its underlying technology layers. It is the obvious term to use when referring to 'what' the ASTRID system actually does. It hints at the fact that ASTRID does inference, which in itself is already a cognitive process.

But when we talk about 'Deep' Inference, in the ASTRID system, we point at the capability to do inference both widely across many knowledge domains and deeply into many layers of related knowledge. This gives the ASTRID system the capability to find any relevant concepts that might lead to new insights.

 

Where humans are quite limited in the amount of information that we can infer at once, the ASTRID system is virtually unlimited in this regard. The amount of knowledge is only limited by storage space and the depth of the inference task is only limited by available CPU power. The basic ASTRID system runs on very lightweight hardware, so both limitations are virtually non-existent.

Our Deep Inference Engine is currently being thoroughly tested and optimized, in the process getting additional inference models to make it even more powerful than it is already. The Deep Inference Engine will be available later this year, in the next ASTRID iteration /BRAIN 6.

© 2021 MIND|CONSTRUCT  
Combining Knowledge Domains
Image: PixaBay 
MODULE 1.2
Combining Knowledge Domains
 

Combining Knowledge Domains is nothing special. We humans do it every moment of the day without even noticing it. But when a machine acquires this capability, we are instantly jumping from 'narrow' Artificial Intelligence to the broad spectrum of what Artificial General Intelligence stands for. In the current world of Deep Learning this is unheard of and still a very distant dream, as Deep Learning is unable to combine seemingly unrelated information into a coherent complex world model.

And then there is the ASTRID system: Capable of not only doing exactly that, but in addition use those combined knowledge spaces to be able to infer analogous concepts. This means it can find things that are only a bit similar to something else but just enough to be useful.

 

The ASTRID system is capable of learning much faster than a human can. When we scale the system to larger server stacks, instead of the basic hardware it needs for normal operation, the system is capable of learning complex information thousands of times faster than humans. Training the system in just one specific knowledge domain takes only hours, days at most.

This is what Super Intelligence really looks like: A machine that is capable of learning everything we know in a short fragment of the time that we need to do that. Subsequently, it has the capacity to use much larger amounts of information to infer the correct (or best) answer to the question at hand.

© 2021 MIND|CONSTRUCT